All right. Good afternoon, and thank you for joining us at the Cloudflare session at the GS Communopia and Technology Conference. I'm Gabriela Borges. I cover the emerging software vector here at Goldman, and I'm delighted to have on stage with me, Thomas Seifert, CFO of Cloudflare. Thank you for your time. Well, thanks for having us. Pleasure being here. So, Thomas, you spoke very consistently on joining Cloudflare and looking at the architecture and continuing to be positively amazed by the amount of technology. Yeah. and the amount of functionality that the architecture can support. Yeah. So I want to frame this question as a little bit of an R&D priorities question, which is: There is so much TAM in the wave one, wave two, and wave three products that the company can go after. You'll have a reputation for being incredibly innovative at the leading edge. Yes. How do you think about R&D priorities in a way that allows you to maintain that innovation and also be focused on the opportunities that can move the needle? This seems to be a theme. This is the second time I asked this question today. I had it in one of the one-on-ones, and it's actually a very timely question because we're just about to kick off our planning session budget for next year and the next three-year plan. Actually, what Phil and I did this morning before we came here, we were sitting down with Michelle, working on the agenda for the next two days. I think the right answer is there is, it's part science and part art, to be very honest. So like every good company, we throw a lot of data science and methodology at the problem, your TAM analysis and all that good stuff. There's also a significant part of art to it at Cloudflare, and that comes with our founder, with Matthew, to be very honest. The astonishing thing about Cloudflare is how little it pivoted over the last 11 years. If you think about it, you know, it has been a story that continued to evolve, but it stayed true to its core of providing and helping to build a better internet. And everything around that, I mean, has evolved around a network architecture that is impressive. We might come to that in a minute, but Matthew seems to have a keen and deep understanding of what holds the internet together from a technology and from a needs perspective. We started to think about data compliance and data governance way ahead of, you know, before it became an issue, and tried to build products around it, like our Data Localization Suite. We talked about the importance of latency and, you know, minimizing the distance of the eyeballs that connect to the network and what that means in terms of the next wave of innovation. We got it actually wrong in the beginning. We thought it was all about speed, and then it turned out to be data compliance and data governance, but it comes now full swing back when we talk about AI and the need of latency and being in this Goldilocks between what can happen on the device and what can happen on the data center, and we find this optimal place. So it's part art and it's part science. But the surprising thing for me, even after six years now, is once you have a network that is architectured the way the Cloudflare network is, and distributed the way it is, and you have this unique software effect that allows you to run every product on every server, and with that, the complete network surface becomes your degrees of freedom, how you manage cost, capacity, compute, storage, how you shift bandwidth across the planet, is I think what is driving this to a very unique degree. So you mentioned the 2024 leadership planning discussion. Yes. How would you compare and contrast the way you feel going into those planning discussions compared to this time last year? What do you think is different? What are your observations on what's changed in the last year? That's an interesting question. You know, the last couple of years were always exciting. Yeah. We, between COVID and post-pandemic and then macroeconomic issues, I think what stays at its core of all the discussions we had is the network and its innovation capability, and how can we keep the flywheel of new products, new features, and TAM expansion going? I think that has been consistent regardless of what other distractions that happened around us. I think what evolves us is how we evolve the go-to market. I remember when I joined in, boy, this is in 2016, six years ago. So, you know, we truly got only our first enterprise customer in 2016, when Mitt Romney ran for president. Bain Capital truly became the first contracted customer, and that is not too long ago. If you think about, you know, today, this is where our revenue sits, and the largest customers are north of $20 million a year. So this evolution continues because even today, the largest customer is not buying all the products. So there's expansion. So the evolving and the sophistication of the go-to market is a theme that changes color from planning period to planning period, but it becomes a top topic. How we instrument us has changed and evolved. We are a highly instrumented company today. So, you know, I can look literally into what is the margin on a product for a customer in a specific location. So how we instrument the company in terms of understanding the levers to evolve it is thriving, is a thriving theme. This year will be special because we are super excited about AI from a top-line perspective, and that drives a lot of activity. But, you know, there's equal opportunity driving productivity within the company by deploying artificial intelligence to get more efficient. I think this is, that is a step up in terms of difference of discussions for this year's budgeting and three-year planning process, what initiatives we are driving for, what projects we are focusing on. Can you give us an example on the second piece of that? So using AI from a cost optimization or workflow automation perspective. You know, we said we try to lead with examples. So this has been a topic of mine for, and a passion of mine for a very, very long time already, way before I joined Cloudflare. So there's a lot of activities that we drive within the finance team across the company, and, you know, how we look at contracts, automatic revenue recognition across- Right. We met with two companies that we are engaged with for the last two years already in that field. So a lot of work on the finance side, how we get to filings and disclosures faster, you know, what we learn about how we script our earnings calls and the messaging. There's a lot of productivity in the go-to-market side. You know, that is running a project. And then I think there's a significant opportunity putting efficiency in our engineering functions, how we code and how we get to a shippable code faster. So there's a joint project I run with our head of engineering. There's quite a lot of opportunity and a lot of projects. It's, you know, we have to make sure that we pick the right ones and that we are not getting lost in too many, and then pick the ones that have the right leverage. Yeah. Is it fair to say that I think about the comments that you've made before on Rule of 40 and solving for revenue growth plus profitability and with all of the potential that you have on the cost side Yes. Is it fair to say that that could be a sorry way of solving for Rule of 50, Rule of 60 over the medium term? Well, I look at it from a productivity perspective. Okay. So, you know, that helps us to lift significant profit, productivity. And then we can decide, you know, how you deploy whatever you gain, you know, what is going to fall through the bottom line and how—what do we reinvest. And as long as there's an opportunity to grow at a significant rate, you know, the best return on the dollar is making sure we continue to grow. But we also have a very, we have a long-term model. We're committed to achieving, you know, north of 20% operating margin and north of 25% free cash flow margin. So we want to get there, probably early on free cash flow than an operating margin, but that is the trade-off we are going to play. But in the first, the first step really is to lift the productivity potential we have. And that is significant, I think. So, I do want to spend time on the go-to-market, but before I do that t he implications for AI on the top line. Yes. One of the comments you've made is small dollar opportunity today, but the potential is huge. Yeah. So how do you think about, y ou've given the examples of customers using AI products or using wave one products for AI use cases. You've given the examples of R2 for wave two. Yeah. How should we think about that ramp in revenue tied to AI customers or Cloudflare customers using more AI in their workflows? So how we ramp to revenue, and that is, you know, always a little bit of a black box. Fair. If you have step-up functions like this. I think what I learned to admire Cloudflare is, y ou know, I remember, and that's also something I mentioned once already today, one of my first interviews with Matthew was really him helping me understand why this business model is so special, and me getting a deeper look into this uniqueness of how the hardware stack is architected, how software runs. And then there was this comment he made, said, "Thomas, the real idea is never to discourage a byte of data from moving through the network. And once it moves through the network, we figure out a way to wrap value around it." And this is the whole idea of premium, right? That's why we have millions of free customers. It's not so much the idea we'll monetize it at some point in the future and get them to pay us dollars. It's the diversity of data and threat intelligence you harvest, the long tail of the market you consolidate, and then we take the data and negotiate interesting colocation and bandwidth agreements with ISP. So AI, I think, works in a very similar manner. We do not want to discourage a byte of data or a model not to run on our network, and we'll figure out a way to monetize it moving forward. But the really, the first incentive is to get as broad as possible from a use case and from an adoption perspective. And I think this is where we have made a significant amount of progress over the last couple of months. You know, the idea that you have this network that found this Goldilocks zone between data centers and devices and being right in the middle. A network that was never really designed to deliver content. It has to deliver content in order to get high performance, security and performance products. So we do so much compute already at the edge of our network today for inspection, decryption, encryption, compression, decompression, so there's so much compute performance on there. If you look at a product like our bot mitigation product, it's an AI product that has been running at the edge of an inference product that has been running at the edge of our network for a while, so we know what that means. So the structure, the architecture of the network is destined to perform there, and now it's getting the use cases on the network, and then we'll find a way to monetize it. But coming back to how I started, never discourage a byte of data from moving through the network, I think is the first step that we have to get right. Have there been observations from either your customer base or your telecoms partners on what the potential limitations of the networking experience that you have today could be? In other words, you're already talking about the amount of compute intensity that has to occur at the co-location zones. Is there a scenario where the step function of that changes in such a way that the network requires more CapEx spend, or requires a slightly different approach to investment, to be able to take Cloudflare to the next level of compute processing, for example, for AI? We don't think so. And you know, this is one of these truly unique competitive modes, I think, of the architecture, is how it scales. So today, every server we add, regardless of where we add it in the world, allows us to increase the efficiency of the network. It allows you to increase the surface of the network, and then you have more degrees of freedom, how you push compute and storage and bandwidth around the world. You know, you might be in New York, and Goldman Sachs is consuming too much of our capacity, so we offload three customers to the next site. Or, you know, our network is underutilized in Tokyo because people sleep, and we offload compute or storage tasks to the part of the network that is underutilized. We think how we scale AI on that network will not interfere with that. We keep the elasticity of the network, and it's literally every server we have today has a slot where you can plug in a GPU card if you need to. So the scalability and the elasticity of the network is not changing with the workloads that we might be adding. And, you know, we think there is more opportunity for us in inference than there is in training. So that plays well into how the network is architected today. So if you look at the simulations we have done in terms of CapEx, CapEx needs moving forward, we think the efficiency. You know, if you go back four years, our network CapEx ratio and the amount of dollars we spend in relationship to revenue has come down, has come down from the high teens to the very low teens. Absolutely. And we think that it's going to continue to scale. We have no indication today that it's gonna break, just the opposite. So, you know, it's one of the unique things about the network. Well, there are two parts to the go-to-market that I want to spend time on. The first is to your point on winning your first enterprise customer back in 2016, and now the movement that you have upmarket. Talk to us about some of the progress that you're making with your new Vice President of Revenue in being able to scale your enterprise go-to-market, and particularly getting the synchronization right of iterating the product to the enterprise grade, and making the decision makers or building relationships with the decision makers at the enterprise simultaneously. Well, it's a journey. You know, it's not like that, a nd that journey has already started to a certain degree, you know, before Mark came. I mean, we were at a billion-dollar run rate already when he came and, you know. But it needs third walls. The customers are getting bigger in terms of ACV. They're getting more complex in terms of the products they buy. And this journey from what we call Act One to Act Two to Act Three requires that we also address, as you said, rightfully, different personas within customers. You have different responsible people for budgets. And we are also moving up the stack in terms of decision makers. Especially in this world now, where you have macroeconomic headwinds, people like my position becomes more important in the decision-making process of how dollars get allocated. So this is not only a matter of adjusting the sales motion, it's also a matter of how we adjust our messaging from a marketing and branding and campaign perspective. We made a change on the sales side that got a lot of attention, but we, you know, we also upgraded our capabilities on the marketing side, actually ahead of that already, and Brent joined us from companies like AWS and F5 before. And so that needs, in conjunction, needs to move up. I think we have a very unique value proposition, especially when it comes to t hreat analytics and, you know, what do we see and how do we interpret the world? We've made a big step forward in making use of the data we see through the acquisition of a company called Area 1, an email product. We acquired a lot of threat intelligence, skill sets, people who have been interpreting threat analytics and have been dealing with threats for a long time. So how we monetize and how we deploy threat analytics in our products and how we sell it even as a separate offering has increased. We published our first phishing report in this quarter. So this allows us to also retarget the personas that we need to get to in a very unique way. And I think it's part now also of the reason why our journey up the enterprise tech has been rather successful. So if you look at the pipeline today, there are probably more large deals in the pipeline than there ever were, and they are a result of this broader product portfolio, different go-to-market motion, talking to the right people, and having more to talk about, in a way, especially from a threat analytics perspective, that has been helping. So it sounds like the threat analytics piece of this gives you more credibility with security buyers and more large organizations. Is that a fair way to frame it? I think it opens doors easier. You know, you have a topic to talk about. You know, it started already last year when Russia invaded the Ukraine, and the war started, and the intelligence we saw, what we saw on the ground, that helped us to up-level messaging and it has evolved, it evolved from there. Yeah. Are there any other observations that you'd call out in terms of how the messaging is changing? We can certainly see some of it from the outside in, but curious how you'd describe it. I think it's more relevant. I think it's more acute to the needs. It addresses... It keeps in mind what the people we target need to talk about, you know? How can you up-level the information a CIO has at a customer when that CIO has to talk to his board and his constituency in terms of, you know, what messages he or she has to deliver? As you've refined the messaging and as you've moved up market, I'm curious how you see Microsoft as a competitor, particularly given they are certainly talking more about AI use cases and introducing more Secure Access Service Edge-type products. Yeah, and, you know, talked about SASE. And, you know, Microsoft has been a, a big partner, a long partner, a valued partner. They were an investor in Cloudflare in, in the, the early phases. Them stepping into the SASE space gives credibility to a market, so that, that is a good thing for us. But it also is at the same time, you know, we take them serious as a competitor, and we see the world a little bit different. One comes from this special place we have. So yeah, our, our place in the network that we operate, this neutral ground, being able to, to, to connect many clouds, I think makes us-- gives us a very unique perspective. How we tie security and networking together is a very unique capability we have that we take full advantage from. You know, our first Act 1 product, you know, protected the external-facing infrastructure. Now, the Act 2 products target the same in the internal-facing properties of networks. So how we tie things together puts us in a very unique perspective that I think allows us to compete really well against everybody in that space. I'm curious if your salespeople have seen a change in how customers think about Microsoft on the ground, or if it's mostly just us from the outside and as industry analyst? I think it's too early to see any big changes. But, you know, if you look at the... I think it helps us to drive an even more differentiated messaging, if at all. You know, this unique place you have of tying network and security together, what we see externally and internally facing this one control plane, the amount of threat intelligence that goes into the product sitting in front of enterprise customers, customers, but a long, long tail of free customers. You know, and some of these free customers are important people that, you know, journalists that sit in interesting places of the world that get attacked by nation-sponsored, attacked by civil society organizations, who sit in front of pretty much all most of the election networks in the United States. We covered every election, presidential election campaign from A to Z, from Biden to Trump and anything in between. So we have a uniqueness about how networking and security comes together that puts us in a good position to compete, even against the Microsoft. One of the strategy pieces of this that I know you spent a lot of time on, is how to be thoughtful on the packaging and bundling of the 50 plus products that fit within the Cloudflare wheelhouse. So would love to hear a little bit on what are your priorities today, and how you think about solutions-based selling and organizing some of these products to make it easier for your sales folks to sell across new AI use cases as they come up market, across security use cases. What are some of the things top of mind on the product and bundling side? Well, you know, you talk about the changes, or changes is probably the wrong word. The dilution that needs to happen on the go-to-market side, bundling is a big part of it. And bundling is not a matter of, you know, bundling within one act. You know, how do you bundle zero trust products? But how do we bundle across the different acts? That, that I think is, you know, how do you use their application security products and the position we have, and how do you bundle that within your offerings? That is where we spend a lot of time trying to understand and optimize that. How do you price that? How do you message this? What are the various bundles for various verticals, and how do you compensate? So, you know, bundling, how to bundle, what systems we use in order to choose that, how we price and how we compensate, is probably, you know, one of the larger work streams that is active currently, and that's an important part for next year, getting that right. When you get that right w hat do you think the implications are going to be for the unit economics of your business and net retention? Do you think we could see a step function change in trend as opposed to just a continuation of trend? Wait, you know, independent of bundling, we have done a really good job, you know, improving unit economics across the board. So we are really well instrumented when it comes to managing unit economics from an offerings, from a revenue, and from a cost perspective. I think the bigger opportunity, the biggest opportunities are less on the unit economic side. Is it going to help us accelerate our expansion metrics? You know, when you have a customer that buys $20 million of product, and that customer is not buying all the products you have, really understanding the dynamics of getting that customer to expand in terms of speed and opportunity, I think is that is where the big uplift is for us. So I would expect, you know, the biggest impact is gonna be on our DNR metrics. And that is where you should see improvement. Then moving forward, whether it's gonna be step function wise, I don't know. I don't want to make that. One step at a time. Commitment, but one step at a time, but it will go up and expand, for sure. Yeah. I'll take the opportunity to ask you about the demand environment. Yeah. The comments from the last earnings call talked about stability. Yeah. Sentiment no longer worsening, sales cycles improving relative to 1Q. Yeah. What do you think has changed over the last three to four months? When you look at the underlying data that you use to forecast any additional observations that you could share with us based on the last two months? Yeah. It's moving along. I cannot say that it improved in a material way. There's some pockets in the world where things are looking better, and there are parts in Europe where things have not gotten better, or there's some parts in Europe where it's deteriorating. So I think on average, it's just moving along. You know, we are also in a very unique position. We've talked about this internally for a while. As you know, when you sell a software product like an ERP system, the sales cycles are really, really long. You know, you implement an ERP product two years, you put it on hold, it has significant implications from a customer perspective in terms of opportunity cost. Implementing a Cloudflare product goes really, really fast, right? So, you know, if you put your entire workload behind us, it takes five minutes. If Goldman comes and says, "We need to onboard our complete network infrastructure," that takes a couple of hours. So putting us on hold and restarting us goes rather quickly, and that's why we think we see changes in macro environments faster than others. Because putting a Cloudflare project on hold, the opportunity costs are not high, right? You can decide tomorrow to restart, and then you're done a couple of hours later. That's why we think, you know, we see—we saw this earlier, and we are probably experiencing changes in sales cycles more directly and more immediately than others. So that is at least our interpretation of the data. From a macro perspective, you know, across everything, we have not seen significant changes, neither to the worst, but also not to the better in the third quarter. You detailed some of the pockets of weakness. Yeah. What are some of the pockets of strength? Parts of Asia, Australia, New Zealand, Japan, South Korea are running well. North America has stabilized, good momentum in Middle America. I would say in pockets of South America, Europe, not so much. How about if you were to cut the data from a large versus small or a vertical or product standpoint? For us, not much different from a vertical perspective. There's no vertical that is exceptionally strong or is exceptionally weak. More difficult expansion with larger customers, easier on new logos, right, regardless of size. Yeah. Let me pause and go to the audience. Please. A little bit. Yeah. Can you talk a little bit about how Act Three will play into just—thanks. How Act Three will play into some of these questions in terms of will it make the upsells land and expand, will it have potentially less friction in the process for your sales force? Is it a bigger learning leap for them and therefore there'll be a productivity issue? It— Will it be an easier sale for certain verticals than others? How should we sort of expect this to look? How do you— And how do you hope that it should look as we get further in that Act Two to Act Three journey? So it's exciting from an opportunity perspective, because the potential lever is really high. Especially if you look at the combination of everything we talked about from an AI workload perspective, the combination of storage at the edge in order to run inference products is crucial. So we think there are certain things that have been in flight for a while, and now they seem to come together really well for certain use cases. And I think AI is gonna be one of the use cases that will demonstrate more than anything else the benefits of this network and where we reside. And in that conjunction, the interplay of compute and storage performance and workers as a tool to deploy code at the edge of our network is gonna shine. You know, I'm not gonna—I don't want to lean in too much today, but those of you who have been following us for a while know that birthday week is coming up. That's, you know, in two weeks we celebrate our twelfth birthday now. And that is normally a period where we just try to overwhelm everybody with new product launches and features. And I think that it will be all centered around that. So moving forward, I think it will be a tailwind for the Act Three products, to be very honest. From a use case perspective, we will be very hesitant to commit dollars behind it. As I said before, the unique thing about us is that we think about the value of the byte in terms of what it can deliver from a non-monetary perspective first, before we think about monetization. But I think it will shine with the new workloads. There was a little bit of a discussion at the analyst day about periods of time in Cloudflare's history, where you focus more on new product introduction versus periods of time where you focus more on taking the existing product set and leveling it up. Yeah. Help us understand how 2024 fits relative to that cadence, given all of your comments on the new products coming out the pipeline, the AI-driven use cases, et cetera. You know, the COVID was a good example of that, where you—where there was one idea, we were still pushing it, but we took the foot of the accelerator, and we was pushing the network even further. Let's say we are in locations today, we in some cities, we are in multiple locations. Let's get into buildings. That was a... And then COVID came, and nobody wanted to be in buildings anymore, so we decelerated for a while. I think 2024 looks like it will be a product innovation year. There is so much excitement around some of the things that are in the pipeline and AI is one of them. If I see how, you know, the in preparation of this budgeting and three-year planning period, there is this bottom-up approach of, you know, who needs money for what and what are all the ideas and where's the competition coming from for this one dollar that we have to spend. There's a lot of excitement in terms of new products and features. So I think 2024 will be one of those years. Is there a scenario where, given your comments on the initial phase is not focused on monetization, it's focused on adoption and engagement and technology innovation? Yeah. Is there a scenario where the pace of investment goes up and margins go down before revenue growth accelerates? No, I think that is exactly why we push adoption first. This allows us to learn. You know, we talked about a lot of the benefits of the business model. One benefit we have not been talking about is that because we see so much data and use cases, we are always—we've always been in a position to invest behind demand and not ahead of it, right? So you—with the use cases and the early adoption of technology, we get a good feeling for what we might need. And we are not in a position where we need to buy 1,000 GPU cards, hoping that they would find capacity. So we are pushing our current infrastructure to the limit in terms of what it can digest, and then we learn of where we might need to rebalance the network or not. This is true for the network and the capacity, regardless of, y ou know, when we started with storage and R2, it was the same topic. You know, we pushed the current network to the limit. We found a way to offload peak demand. And once you have thousands of use cases, you have a really good understanding of how the need is going to play out, and you invest behind the demand curve. On the GPU side, we'll follow a very similar model, and we have been following it for the last 12 months already. You have to keep in mind that this is one really interesting interview with Matthew, where I said: You know, we, when I got started, we, we talked about us being an AI company, and I stopped talking about AI because everybody rolled his eyes, and I, I learned that this is not a good thing. We've been running our own inference models at the edge of our network for 12 years, and maybe it was more about machine learning than generative AI, but we, we really understand well what it takes to run inference models at the edge of our network. The amount of skills we have in-house already is pretty unique, and that's why we feel rather good about understanding what it takes to build out the network without coming in a situation where we have to over-invest or invest ahead of a demand that might or might not appear. Excellent. Yeah. Thomas, thank you for your time. We'll leave it there. Always a pleasure. Thank you. Yes.